Contrib Mineral Petrol (2000) 138: 249±264 Ó Springer-Verlag 2000

Peter W. Reiners á Paul E. Hammond Juliet M. McKenna á Robert A. Duncan Young basalts of the central Washington Cascades, ¯ux melting of the mantle, and trace element signatures of primary arc magmas

Received: 2 March 1999 / Accepted: 18 August 1999

Abstract Basaltic lavas from the Three Sisters and correlated with the amount of SC added to it. Distinctive Dalles Lakes were erupted from two isolated vents in the CAB and OIB-like trace element compositions are best central Washington Cascades at 370±400 ka and explained by a ¯ux-melting model in which dF/dSC 2.2 Ma, respectively, and have distinct trace element decreases with increasing F, consistent with isenthalpic compositions that exemplify an important and poorly (heat-balanced) melting. In the context of this model, understood feature of arc basalts. The Three Sisters la- CAB trace element signatures simply re¯ect large de- vas are calc-alkaline basalts (CAB) with trace element grees of melting of strongly SC-¯uxed peridotite along compositions typical of most arc magmas: high ratios of relatively low dF/dSC melting trends, consistent with large-ion-lithophile to high-®eld-strength elements derivation from relatively cold mantle. Under other (LILE/HFSE), and strong negative Nb and Ta anoma- conditions (i.e., small degrees of melting or large degrees lies. In contrast, the Dalles Lakes lavas have relatively of melting of weakly SC-¯uxed peridotite [high dF/ low LILE/HFSE and no Nb or Ta anomalies, similar to dSC]), either OIB- or MORB (mid-ocean ridge basalt)- ocean-island basalts (OIB). Nearly all Washington like compositions are produced. Trace element and iso- Cascade basalts with high to moderate incompatible topic compositions of Washington Cascade basalts are element concentrations show this CAB or OIB-like easily modeled by a correlation between SC and F across compositional distinction, and there is pronounced a range of mantle temperatures. This implies that the divergence between the two magma types with a large dominant cause of arc magmatism in this region is ¯ux compositional gap between them. We show that this melting of the mantle wedge. trace element distinction can be easily explained by a simple model of ¯ux-melting of the mantle wedge by a ¯uid-rich subduction component (SC), in which the degree of melting (F) of the peridotite source is Introduction

Arc basalts are typically enriched in LILE and other ¯uid P.W. Reiners (&)1 mobile elements (e.g., Cs, Rb, Ba, Sr, K, U, Th, LREE) Division of Geological and Planetary Sciences, relative to most HFSE (especially Nb and Ta, and to MS 100-23, California Institute of Technology, Pasadena, CA 91125, USA lesser degrees Hf, Zr, and Ti). This characteristic is gen- e-mail: [email protected] erally thought to be caused by addition of a LILE-rich ¯uid or melt derived from a subducted slab to a mantle P.E. Hammond Department of Geology, Portland State University, wedge source (e.g., Gill 1981; Tatsumi et al. 1986; Portland, OR 97207, USA Hawkesworth et al. 1993, 1994; Stolper and Newman J.M. McKenna 1994). Basalts displaying such trace element features are Montgomery Watson Americas, Inc., Pasadena, CA 91101, USA commonly termed CAB (e.g., Leeman et al. 1990; Conrey R.A. Duncan et al. 1997; Bacon et al. 1997). Many arc basalts, how- College of Oceanic and Atmospheric Sciences, ever, especially in the Cascades, have relatively high in- Oregon State University, Corvallis, compatible element concentrations but do not have high OR 97331, USA LILE/HFSE or negative Nb and Ta anomalies. Because Present address: these trace element characteristics are common to many 1 Department of Geology, Washington State University, Pullman, ocean-island basalts, these have been termed intraplate-, WA 99164, USA or OIB-like basalts (e.g., Luhr et al. 1989, Luhr 1997; Editorial responsibility: T.L. Grove Leeman et al. 1990; Conrey et al. 1997; Bacon et al. 1997; 250 Ma rquez et al. 1999), even though the arc in which they 116 49N are found may have no obvious connection to intraplate- Mt. Baker ° W or plume-related tectonic settings or processes. Glacier Peak Here we report age and chemical results from young N lavas of two small basaltic ®elds in the central Wash- 0 km 200 Mt. Rainier ington Cascades that clearly exemplify the CAB-OIB Mt. St. Goat Rocks compositional distinction. We show that this distinction Helens Mt. Adams 46N Simcoe cannot be explained simply by di€erent amounts of Boring Indian Heaven ¯uid-rich subduction component (SC) from the sub- Lavas Mt. Hood ducting slab in each mantle source. However, the dis- Lahars of Mt. Rainier Basalt of Three Sisters tinct and divergent CAB-OIB array is easily modeled as and Dalles Lakes ¯ux-melting of the mantle, involving a correlation be- Kent Mt. Rainier basaltic andesite tween the amount of SC added to the mantle source and Mt. Rainier andesite the degree of melting of the source. One important im- and dacite plication of this modeling is that OIB-like arc basalts can Tertiary intrusive be produced by the same melting process and from the rocks 47°15' same mantle source as CAB melts, simply by relatively Puyallup Enumclaw low-degrees of melting of a mantle region that has had a comparatively small amount of slab component added to it. Thus, OIB-like trace element compositions do not Orting Dalles necessarily indicate the presence of plume material or Three Sisters Lakes ocean-island-like melting processes under arcs. 47°00'

Samples and methods

The Three Sisters (TS) and Dalles Lakes (DL) basalts were erupted from relatively small and presumably short-lived vents about 30 km north of Mt. Rainier Mt. Rainier (Fig. 1). Our mapping, as well as earlier work (Fischer 46°45' 1970; Hartman 1973; Hammond 1980), indicates that 0510 15 20 the TS and DL basalts were erupted subaerially through km after Luedke & Smith (1982) older andesitic porphyry and volcaniclastic deposits. 122°15' 122°00' 121°45' 121°30' Each vent produced approximately 4±10 lava ¯ows with a similar phenocryst content of olivine, and rare pla- Fig. 1 Generalized geologic map of the central Washington Cascades near Mt. Rainier (after Luedke and Smith 1982; gioclase and/or clinopyroxene. The TS and DL basalts Hammond 1998) showing the locations and approximate areal are the only known volcanic rocks of the High Cascades distribution of the Three Sisters and Dalles Lakes basalts. Shown in (late Cenozoic to recent) between Mt. Rainier and the inset are the locations of other Washington and northern Glacier Peak (Hammond 1980; Luedke and Smith 1982), Oregon volcanoes (triangles) and selected volcanic ®elds discussed in this paper (squares). The Three Sisters and Dalles Lakes basalts a distance of nearly 150 km. Their apparently primitive were originally described by Fischer (1970) and Hartman (1973), compositions provide insights into primary magma who named them the Basalts of Canyon Creek and Dalles Ridge, compositions and mantle-melting processes in this part respectively. For stratigraphic clarity (because of the abundance of of the Cascade arc. ``Canyon Creek'' names in the Cascades, and because we reserve the name ``Dalles Ridge'' for one of the thick, older volcanic units We collected and analyzed 7 TS and 11 DL samples that composes most of the ridges in the area) we suggest the new for major- and trace-element compositions (Table 1). names Analyses were performed at Washington State Univer- sity and CHEMEX Labs (Vancouver, B.C.) using standard XRF methods for major elements and ICP-MS than the Three Sisters basalt (Table 1). Single-step fusion for trace elements. Three (including two ¯ows and one ages for two samples (365 ‹ 50 and 385 ‹ 71 ka) and dike) TS samples and one DL ¯ow sample were dated by plateau and isochron ages of one sample (398 ‹ 15 and whole-rock, single-step, total-fusion 40Ar/39Ar methods, 377 ‹ 42 ka, respectively) indicate an age of about 370± and one TS sample was also dated by incremental 400 ka for the Three Sisters (Table 1). The single-step heating 40Ar/39Ar methods, at Oregon State University. fusion procedure of 96PRTS6 produced a very low ra- diogenic Ar yield and large analytical uncertainty that precludes con®dent age assignment, but it is concordant Results with the results of the other samples. High MgO (8±10 wt%), Mg#(60±70), and compatible 40Ar/39Ar ages indicate that the Dalles Lakes basalt was trace element contents indicate that the DL and TS lavas erupted at 2.2 ‹ 0.05 (1r) Ma, and is signi®cantly older are primitive basalts that have experienced little 251

Table 1 40Ar/39Ar ages (Ma) of samples of the Dalles Lakes 96PRDR2 96PRTS1 96PRTS6 96PRTS11 and Three Sisters basalts, cen- tral Washington Cascades. Single-step, total fusion age 2.18 ‹ 0.05 0.385 ‹ 0.071 0.271 ‹ 0.174 0.365 ‹ 0.050 ‹are 1r Radiogenic Ar 89.1% 38.3% 1.30% 74.1% Plateau (2 steps, 80% of gas) 0.398 ‹ 0.015 Isochron (5 steps) 0.377 ‹ 0.042 fractional crystallization (Tables 2, 3). In many other (>52.5%) TS samples show slightly di€erent trace respects, however, major- and trace-element composi- element trends than the rest of the TS suite, with higher tions of DL and TS lavas are distinct: DL lavas are Ni and incompatible element concentrations. DL lavas alkalic basalt whereas TS lavas are subalkalic basalts also show interesting intrasuite trends, particularly and basaltic andesites (Fig. 2). DL lavas also have high correlations between Ni and incompatible elements such TiO2 and Fe2O3, and low SiO2 relative to the TS lavas as Ba, La, and Sr (Tables 2, 3). Overall, however, the (Fig. 2, Table 1). In detail, there are subtle but notable most remarkable compositional distinctions are compositional variations among samples from each contrasting ratios of LILE and other ¯uid-mobile suite. In particular, most of the relatively high SiO2 elements (e.g., Cs, Rb, Ba, Th, U, Sr), and HFSE (e.g.,

Table 2 Compositions of samples of the Dalles Lakes basalt, central Washington Cascades

96PRDR2 PHA92262 PHA92264 PHA92265 PHA92266 PHA92267 PHA92268

SiO2 47.60 48.90 49.17 50.14 49.52 47.98 48.53 TiO2 1.87 1.41 1.53 1.49 1.51 1.22 1.79 Al2O3 16.05 16.25 15.67 16.20 16.06 16.70 15.27 Fe2O3 10.6 11.52 10.68 11.34 11.34 12.21 10.17 CaO 8.19 8.88 8.69 8.85 8.74 9.06 8.38 MgO 8.06 8.01 8.37 8.07 8.13 9.03 8.3 MnO 0.16 0.16 0.16 0.16 0.16 0.18 0.15 K2O 1.23 0.70 0.91 0.83 0.82 0.35 1.23 Na2O 3.22 3.49 3.56 3.76 3.47 3.12 3.59 P2O5 0.44 0.31 0.32 0.29 0.29 0.17 0.44 LOI 0.90

Total 98.37 99.63 99.06 101.13 100.04 100.02 97.85 Ba 359 171 207 170 192 94 297 Ce 55 34 45 39 28 29 56 Cs 0.3 Co 44.5 Cr 342 267 273 265 270 247 270 Cu 45 33 52 53 48 17 42 Dy 4.7 Er 2.4 Eu 1.8 Gd 5.4 Ga 20 20 18 16 18 14 18 Hf 4 Ho 0.9 La 25 3 13 6 11 22 21 Lu 0.3 Nd 26 Ni 195 139 153 144 147 135 162 Nb 29 16.6 21.4 17.8 19.5 8.3 31.9 Pr 7.1 Rb 20 6 14 14 12 6 19 Sc 29 28 28 32 36 30 Sm 5.7 Sr 589 396 457 422 421 302 574 Ta 2.5 Tb 0.9 Th 2 2 2 3 Tm 0.4 U 0.5 V 175 178 182 176 185 179 181 Yb 2.6 Y24242324242423 Zn 90 89 87 88 87 82 82 Zr 187 131 153 140 144 100 191 252

Table 3 Compositions of samples of the Three Sisters basalt, central Washington Cascades

96PRTS1 96PRTS6 96PRTS11 PHA97057 PHA97058 PHA97059 PHA97060 PHA97061 PHA97062 PHA97064 PHA97065

SiO2 50.98 50.56 52.51 52.60 52.38 51.93 50.12 50.78 53.32 53.53 54.42 TiO2 1.15 1.09 1.25 1.15 1.12 1.18 1.19 1.17 1.16 1.18 1.26 Al2O3 15.67 15.55 15.34 16.35 15.80 16.36 16.58 16.63 15.61 15.83 15.56 Fe2O3 9.85 9.36 8.90 9.29 9.17 9.47 10.16 9.78 8.55 8.32 7.99 CaO 8.70 8.33 7.65 8.70 8.90 8.41 8.23 8.81 7.67 7.75 7.61 MgO 8.83 9.57 9.07 8.22 8.19 8.36 9.94 9.23 9.05 8.64 8.54 MnO 0.14 0.14 0.13 0.14 0.14 0.15 0.16 0.15 0.13 0.13 0.12 K2O 1.2 1.02 1.43 1.16 1.19 1.26 0.71 0.91 1.35 1.37 1.45 Na2O 2.9 2.82 3.21 3.2 3.19 3.07 2.73 3.01 3.43 3.49 3.65 P2O5 0.29 0.28 0.36 0.30 0.29 0.30 0.25 0.30 0.33 0.35 0.36 LOI 0.39 0.51 0.0

Total 100.15 99.3 99.92 101.11 100.37 100.49 100.06 100.76 100.60 100.60 100.96 Ba 369 360 450 348 350 353 297 379 364 396 407 Ce 55.5 56 59.5 58 63 71 42 62 68 66 57 Cs 0.5 0.6 1.1 Co 41 41.5 38 Cr 411 479 479 314 309 332 392 356 388 359 372 Cu 20 30 35 6 9 2 13 7 15 25 16 Dy 4.2 3.7 3.5 Er 2.5 2.4 2.1 Eu 1.6 1.6 1.6 Gd 4.3 4.6 4.5 Ga 19 20 20 21 17 19 21 16 19 17 21 Hf544 Ho 0.8 0.8 0.8 La 23 24.5 26 26 23 28 35 23 26 24 20 Lu 0.4 0.4 0.3 Nd 27 26.5 27.5 Ni 85 130 190 52 54 64 94 58 185 158 168 Nb 7 8 13 6.9 7.1 7.9 8.3 7 11.8 11.1 13.7 Pr 7.3 7.5 7.9 Rb 22 17.6 30.8 20 19 21 15 6 24 20 26 Sc 29 30 30 27 31 24 22 20 Sm 5.3 5.2 6.1 Sr 728 718 763 720 743 711 598 787 752 809 762 Ta 0.5 0.5 1.0 Tb 0.7 0.7 0.7 Th34425567597 Tm 0.4 0.4 0.2 U 1.0 1.5 1.5 V 150 135 135 182 169 171 176 181 164 167 181 Yb 2.3 2.1 1.8 Y 22.5 21.5 19.5 22 23 23 25 24 20 20 19 Zn 70 75 90 73 72 82 83 74 72 72 77 Zr 166 150 157 176 177 178 172 187 173 174 178

Nb, Ta, Zr, Ti). These di€erences are evident in spider LILE contents that are not correlated with Nb content. diagrams, which show that the TS lavas have strong These lavas also have generally high Ba, Th, Ba/Nb, and negative Nb and Ta (and to a lesser degree Ti) anoma- Ba/Zr, and low Nb/Zr. They also typically have rela- lies, whereas the DL lavas do not (Fig. 3). tively high Cs/K, La/Nb, SiO2, and K2O, and low Fe2O3 Strongly contrasting LILE/HFSE compositions dis- and TiO2 (Leeman et al. 1990; Bacon et al. 1997; Conrey tinguish not only the DL and TS basalts but also many et al. 1997). basalts from the southern Washington Cascades A clearly separate group of basalts, consisting of DL (Fig. 4), and have been used to de®ne two distinct types and other OIB-like lavas is best characterized by a wide of primary magmas (Leeman et al. 1990; Bacon et al. range of HFSE, which is well correlated with LILE con- 1997; Conrey et al. 1997). High LILE/HFSE and nega- tent, although LILE are generally at lower concentrations tive Nb and Ta anomalies of the TS lavas are typical of (for a given HFSE concentration, e.g., Fig. 4). The OIB- what is often called the arc basalt, or CAB. CAB lavas, like lavas also have low Ba/Nb and a wide range of Nb/Zr, including many from Indian Heaven, Mt. Rainier, Mt. correlated with Ba/Zr. These have been termed intraplate, Adams, and scattered localities in southern Washington or OIB (Leeman et al. 1990; Bacon et al. 1997; Conrey and northern Oregon (Fig. 4), are typi®ed by low HFSE et al. 1997). Besides Dalles Lakes, these lavas are common (e.g., <20 ppm Nb), and variable but generally high on Mt. St. Helens, Simcoe, and Mt. Adams. 253

7.0 the ages and compositions of the TS and DL basalts make it unlikely that they are related to the magmatic system of Mt. Rainier. OIB-like trace element compo- alkalic 6.0 Mt.Rainier sitions of the DL basalts are unlike any known Rainier %) Cascade CAB sub-alkalic andesites sample, and the 2.2 Ma age of the DL vent predates the wt. ( earliest known Rainier eruption at about 500 ka (Sisson O 5.0 Dalles Lakes and Lanphere 1997). Although trace element composi- Three Sisters + K tions of the TS lavas are similar to those of the basaltic O 22 Mt. St. Helens andesites at Echo and Observation Rocks, on the north Na 4.0 Indian Heaven Simcoe side of Mt. Rainier (McKenna 1994), the TS lavas are Mt. Adams more ma®c (8±10 versus 6±7 wt% MgO), and much Mt. Rainier BA older than Echo and Observation Rocks (105 ka, from 3.0 46 51 56 61 K/Ar dating by Lanphere and Sisson 1995). SiO2 (wt.%) Contrasting patterns of LILE and HFSE concentra- tions among the DL and TS basalts clearly exemplify the Fig. 2 Total alkalis (Na2O+K2O) versus SiO2 (wt%). Data from volcanoes in central and southern Washington other than Three distinction between CAB and OIB-like lavas of the Sisters (TS) and Dalles Lakes (DL) are from Leeman et al. (1990), Cascades (Figs. 3, 4). At high or moderate concentra- Conrey et al. (1997), Bacon et al. (1997), and McKenna (1994). tions of incompatible elements (e.g., high Ba/Nb or Nb/ Cascade CAB ®eld includes samples in Conrey et al. (1997) Zr), trace element di€erences between CAB and OIB- classi®ed as either CAB or high-K CAB. Alkalic-subalkalic basalt dividing line is from Miyashiro (1978) like lavas are not a matter of degree. There appears to be markedly distinct ®elds that diverge from a common region at low incompatible element concentrations (e.g., 100 low Ba/Nb and low Nb/Zr). This distinct divergence of the CAB and OIB-like magmas and the lack of magmas with compositions intermediate between the two groups 10

ve mantle ve may provide an important constraint on sources and processes involved in arc magma genesis.

1.0 Dalles Lakes Three Sisters

Sample/primiti Trace element modeling

0.1 To investigate the potential signi®cance of the distinct Cs Rb Ba Th U Nb Ta K La Ce Sr Nd Sm Zr Hf Eu Ti Gd Yb divergence of CAB and OIB-like compositions, we Fig. 3 Trace element compositions of DL sample 96PRDR2 and modeled the trace element e€ects of partial melting of average of three TS samples (96PRTS1, 6, and 11), normalized to peridotite mixed with (i.e. metasomatized by) varying primitive mantle (Sun and McDonough 1989). Note enrichment of TS in Cs, Rb, Ba, Th, and U, and depletion in Nb, Ta, and Ti, amounts of a LILE-rich SC derived from a subducting relative to DL slab. In most of our modeling we used a mantle peridotite composition and an SC composition (Fig. 5; Table 4) similar to those used in previous studies (Stol- Although not as obvious as the CAB and OIB-like per and Newman 1994; Borg et al. 1997), but with some basalts in Fig. 4, a third type of basalt, the low-K tho- modi®cations. Abundances of several elements have leiite, high-alumina olivine tholeiite, or MORB-like ba- been adjusted to provide good ®ts to ¯ux-melting trends salt, also occurs in the Cascades. MORB-like basalts in trace element space, as presented below. In detail, the typically have low incompatible element concentrations, model peridotite composition we used in most of the falling near the origin in most of the plots in Fig. 4. The model examples is slightly more enriched in incompati- MORB-like basalts are found throughout the Cascades, ble elements than that of Stolper and Newman's (1994) and those plotted in Fig. 4 include the low Ba, Nb, and and Borg et al.'s (1997) MORB source. However, the Th lavas from the Mt. Adams and Indian Heaven model peridotite is more depleted than Borg et al.'s OIB groups (Leeman et al. 1990; Bacon et al. 1997). source, except for Nb and Ta, which are slightly more enriched (1.50 ppm Nb, 0.14 ppm Ta, compared with 1.13 and 0.10 ppm, respectively, in the OIB source). The Discussion positive Nb and Ta anomalies assumed for this perido- tite source are consistent with positive Nb anomalies The 40Ar/39Ar dates of the DL and TS basalts con®rm observed in unmetasomatized xenoliths from the conti- the ®eld-based suggestions of young eruption ages nental lithospheric mantle in southern British Columbia (Fischer 1970; Hartman 1973; Hammond 1980). They (Sun and Kerrich, 1995), which may be similar to the also demonstrate that late Cenozoic to recent volcanic mantle source under the Cascades. activity in the central Cascades is not restricted to large Furthermore, Ionov and Hofmann (1995) showed stratovolcanoes and their immediate satellite vents. Both that subcontinental mantle xenoliths with small modal 254

Fig. 4 Concentrations of, and 80 ratios involving, LILE (Ba), 800 c Th, and HFSE (Nb, Zr) for the CAB a up to Ba/Nb =170 Three Sisters and Dalles Lakes 60 basalts, as well as basalts from 600 CAB other areas in the central and southern Washington Cas- cades. CAB and OIB lavas OIB 40 400 Ba form distinct and divergent Ba/Nb trends in these plots, primarily caused by the low HFSE and 20 high LILE/HFSE of CAB la- 200 OIB vas. There is a paucity of lavas M with intermediate compositions M at moderate to high concentra- 0 0 tions of incompatible elements. CAB b up to Ba/Zr = 5.2 d MORB-like basalts (near the 3.0 region labeled M) have low 8 OIB concentrations of incompatible CAB elements (and low Ba/Nb, Nb/ 6 Zr, and Ba/Zr) relative to CAB 2.0 and OIB-like lavas. Data from Th Ba/Zr sources listed in Fig. 2 4 OIB

2 1.0 M 0 M 0 10 20 30 40 50 60 0.025 0.075 0.125 0.175 0.225 Nb Nb/Zr

Three Sisters Mt. Adams Indian Heaven Dalles Lakes Rainier bas. and.'s Rainier andesites

Mt. St. Helens Simcoe Cascade CAB

Table 4 Peridotite and subduction component (SC) compositions for Nb and Ta). SC concentrations for La, Ce, and Sm are from used in model. Peridotite concentrations for Cs, Rb, U, K, La, Ce, Borg et al. (1997) (many of which are from Stolper and Newman Hf, and Yb are average of Borg et al. (1997) MORB and OIB 1994), other elements adjusted to ®t trends, but are similar to those sources. Other elements are adjusted to ®t trends in Fig. 7, but are of Borg et al. (1997) and Stolper and Newman (1994) intermediate between Borg et al.'s MORB and OIB values (except

Peridotite SC Borg et al. Borg et al. Borg et al. MORB source OIB source SC

Cs 0.00465 3.5 0.0003 0.009 2 Rb 0.413 120 0.026 0.80 147 Ba 9 1500 0.288 13.1 1371 Th 0.07 16 0.0060 0.16 11 U 0.02 5.3 0.0025 0.028 4 Nb 1.5 8 0.12 1.13 46 Ta 0.14 0.5 0.0075 0.10 3 K 283.85 45,000 27.7 540 70,559 La 0.815 99 0.150 1.48 99 Ce 2.47 297 0.490 4.45 297 Sr 18 3500 5.77 28.7 4681 Nd 1.4 130 0.654 2.97 147 Sm 0.35 35 0.293 1.01 35 Zr 25 500 7.2 26.4 985 Hf 0.4 20 0.20 0.60 27 Eu 0.13 9 0.119 0.26 13 Ti 1000 30,000 996 1980 85,000 Gd 0.45 30 0.467 0.86 39 Yb 0.5 12 0.402 0.56 32 fractions (2±4%) of disseminated amphibole have almost such trace element characteristics may dominate large identical Nb and Ta concentrations and very similar regions of the mantle wedge under subduction zones. overall trace element patterns (Fig. 5). Peridotite with Although Borg et al. (1997) included amphibole in some 255

Fig. 5 Trace element composi- 1000 tions of peridotite and SC sources assumed in this study (bold black lines), compared with sources from Borg et al. 100 SC (Borg et al.) (1997) (dashed lines), and a peridotite xenolith from the Siberian subcontinental mantle. SC (Ionov and Hofmann 1995) 10 ve mantle ve OIB source (Borg et al.)

1.0 peridotite

Sample/primiti enolith amph. x 0.1 MORB source (Borg et al.) 0.01 Cs Rb Ba Th U Nb Ta K La Ce Sr Nd Sm Zr Hf Eu Ti Gd Yb of their models, our model peridotite does not contain We modeled trace element e€ects for several di€erent amphibole, following the modeling of Stolper and New- SC addition and partial melting scenarios. In each case, man (1994), the thermodynamic models of Hirschmann we model melts as mass-weighted aggregates of modal, et al. (1999), and the experimental results of Gaetani and incremental-batch partial melting (in increments of 0.1± Grove (1998). Pargasitic amphibole has been produced 0.4% melting, depending on the model and integrated experimentally in spinel lherzolite in simulated metaso- extent of melting) of a mixed SC peridotite source with matic reactions (e.g., Sen and Dunn 1995), but experi- 55% olivine, 30% orthopyroxene, 10% clinopyroxene, mental temperatures are generally subsolidus and much 2.5% garnet, and 2.5% spinel. Modeling the melting as lower (950±1025 °C) than those assumed (>1225 °C) for non-modal, pure batch, or pure fractional melting does the present study. Other experimental evidence (Dunn not strongly a€ect the results for the elements we exam- et al. 1993; Menner and Dunn 1995; Niida and Green ined. Due to uncertainties in the depth of subarc mantle 1999, and references therein) indicates that at pressures of melting we assumed equal proportions of garnet and 1±3 GPa amphibole breaks down or reacts completely spinel in the peridotite; changing the source to either all out of the modal assemblage at relatively low tempera- garnet or spinel does not signi®cantly a€ect these results. tures near the solidus. For these reasons, we assume that Distribution coecients for all elements were taken from at the relatively high degrees of melting required to re- Borg et al. (1997), and references therein. produce magma compositions in our models (typically Figure 6 illustrates the three di€erent end-member >5%), amphibole would not be a residual phase and scenarios of partial melting of SC peridotite mixtures, would not a€ect the melt trace element patterns. Other for which we modeled the trace element e€ects on melts. trace element features of the DL and TS lavas, such as The ®rst scenario (Fig. 6A) assumes no correlation be- lack of negative Ba anomalies, also do not suggest sig- tween the amount of SC added to the peridotite and the ni®cant amounts of residual amphibole. degree of melting. We modeled the trace element e€ects The model SC composition that we use in most of the of this scenario as progressive melting of individual modeling (Fig. 5; Table 4) is very similar to that of Borg mixed peridotite-SC (metasomatized) sources, with et al. (1997), which is the same as that of Stolper and either 0, 1.5, or 5% SC. Newman (1994) for all elements except those added by The second scenario (Fig. 6B) considered is one in Borg et al.. There are, however, some di€erences. The which the degree of melting (F) of the peridotite is a most extreme di€erences are in the HFSE; the Nb, Ta, linear function of the amount of SC added to the Zr, and Hf concentrations of this study are signi®cantly peridotite. The theoretical basis for a correlation more depleted (by factors of 1.3 to 6) than those in Borg between SC and F is well established. It is based on et al.'s SC. The results of the ¯ux melting models are not numerous experimental and theoretical studies (e.g., highly sensitive to the exact SC composition used, as Hirose and Kawamoto 1995; Gaetani and Grove 1998; long as it is strongly depleted in HFSE, which is widely Hirschmann et al. 1999) showing that addition of a believed to be an important chemical feature of slab- volatile- and/or alkali-rich component to mantle derived ¯uids (e.g., Gill 1981; Tatsumi et al. 1986; peridotite lowers the solidus temperature of the perido- Hawkesworth 1993, 1994; Stolper and Newman 1994; tite, causing, or increasing the extent of peridotite partial Ayers et al.1997). The initial peridotite composition melting. These studies also show that peridotite tem- does, however, strongly in¯uence trace element compo- perature has a strong in¯uence on the slope of the SC-F sitions of the melts. We show several important e€ects of correlation: hotter mantle melts to a larger degree per varying both SC and peridotite initial compositions in increment of SC metasomatism (higher dF/dSC) than the model results. cooler mantle (Fig. 6B). Under isothermal conditions 256

2.5 least up to melt fractions of about 20%, corresponding a no correlation to clinopyroxene-out reactions (Hirschmann et al. 5.0% 2.0 SC 1999). Thus, this second class of model can be consid- ered to represent the e€ects of isothermal ¯ux melting. 1.5 There is also strong geochemical evidence that linear SC- 1.5% SC F correlation approximating isothermal ¯ux melting 1.0 functions (Hirschmann et al. 1999; Gaetani and Grove 1998) can explain trace element variations of arc related 0.5 magma suites (Stolper and Newman 1994). Using the 0% SC isothermal ¯ux melting MELTS models of Hirschmann 0 et al. (1999), which assumes an SC of 45% H2O, trace element e€ects of isothermal ¯ux melting of three dif- b linear (isothermal) 0.8 ferent peridotite temperatures (1225, 1275, and 1375 °C) have been modeled. We calculate dF/dSC for each 0.6 temperature by approximately parameterizing Hi- rschmann et al.'s isothermal ¯ux melting calculations as: 0.4 dF ˆ 3  106e0:012T 1† T=1300°C T=1350°C dSC S&N94

SC (wt.% SC in source) 0.2 T=1250°C The third melting model we examined assumes that F T=1375°C and SC are non-linearly correlated, speci®cally that 0 d2F/dSC2 < 1, or melting productivity decreases as SC c curved (isenthalpic) and F increase (Fig. 6C). This scenario is similar to 0.8 isothermal ¯ux melting because larger amounts of SC cause more extensive melting, and higher initial 0.6 peridotite temperature causes higher dF/dSC. The 0 T0 =1325°C T =1375°C 0 strong curvature of the SC-F trends, however, is pre- 0 T =1275°C 0.4 T =1225°C dicted for the case of isenthalpic (rather than isother- T0 =1425°C mal) ¯ux melting, due to the fact that heat consumed 0.2 by progressive melting decreases the temperature of the peridotite, thus decreasing dF/dSC ratios with in- 0 creasing F and SC (Hirschmann et al. 1999). We have 0 5 10 15 20 modeled the trace element e€ects of isenthalpic ¯ux F (% melting of source) melting for three cases with di€erent initial peridotite Fig. 6 Three potential relationships between extent of subduction temperatures (1225, 1275, and 1375 °C, Fig. 6C). Our component (SC) metasomatism of the peridotite source, and ¯ux melting equations are derived by using the tem- melting (F ) of the mixed peridotite-SC source. A No relationship perature dependence of dF/dSC from Hirschmann between amount of SC added to peridotite and F of mixed-source et al.'s models [i.e., Eq. (1)], and then assuming that melting (i.e., progressive melting of individual, variably metasoma- with progressive melting, peridotite temperature tized sources). We evaluated the trace element predictions of this scenario for three separate sources with 0, 1.5, and 5% SC. B decreases as a function of F, latent heat of fusion (DHf) Linear correlation between SC and F. SC-F trends are taken from and heat capacity (Cp), assumed to be 753 J/g, and Hirschmann et al. (1999) and represent isothermal ¯ux-melting 1.3 J/g K, respectively (Hess 1992; Langmuir et al. trends for peridotite of di€erent temperatures at 1.0 GPa, as a 1992), as: result of progressive H2O added to the peridotite (SC is assumed to comprise 44% H O, e.g., Stolper and Newman 1994). Hotter 2 DHf mantle melts to a larger degree per increment of SC metasomatism T ˆ T0 F 2† C than colder mantle. Dashed line labeled S&N94 is ¯ux melting trend p calculated from compositions of Mariana back-arc lavas by Stolper Substituting (2) into (1) yields: and Newman. C Curved SC-F correlations (d2F/dSC2 < 1). As noted by Hirschmann et al. (1999) such curved SC-F trends are  dF 0:012 T FDHf predicted by isenthalpic ¯ux melting, in which peridotite temper- ˆ 3  106e 0 Cp 3† ature decreases with progressive SC metasomatism and melting. dSC Trends in this model are assumed to pass through the origin (some SC metasomatism is required to initiate melting), and are We apply this equation iteratively to progressively melt a calculated by parameterizing dF/dSC as a function of T in the mixed peridotite-SC source as a function of SC added isothermal trends of Hirschmann et al. (1999), and assuming (in 0.01 wt% increments), correcting the residual trace temperature of the peridotite changes with progressive melting as element concentrations after each increment of melting, T ˆ T ) FDH /C , where T is initial peridotite temperature prior 0 f p 0 adding a new increment of SC, and aggregating the melt to melting, DHf is latent heat of fusion and Cp is heat capacity produced up to 30% melting. (in which the temperature of the peridotite is held con- In both the isothermal and isenthalpic models, we as- stant regardless of the extent of SC addition or melting), sume that F ˆ 0 at 0% SC added to the peridotite at any the SC-F correlation will be approximately linear, at temperature, in contrast to the Hirschmann et al. models 257 (Fig. 6B) in which at 1.0 GPa, peridotite at any temper- sources, and (2) isothermal ¯ux melting. Several im- ature over about 1250 °C has >1% melt present. Our portant conclusions regarding the relative shapes of the assumption is equivalent to an onset of melting requiring melting trends and data ®elds can be easily seen with an addition of a ®nite mass of SC, rather than a peridotite these plots, and the overall results of these examples can source with a pre-existing melt fraction >0%. This is be generalized over a wide range of concentrations and overly simplistic in detail, but changing the model to as- ratios of other trace elements. sume a pre-SC-¯ux melt fraction does not a€ect the im- Partial melting of individual mantle sources that have portant results of our modeling, which are caused by the been metasomatized by di€erent amounts of SC (sce- variations in slopes with temperature and melt fraction. nario illustrated in Fig. 6A) predicts a series of positively Additionally, the 0% pre-SC-¯ux melt fraction condition correlated, slightly concave upward trends (dashed lines) could be considered to suggest that hotter mantle sources in Fig. 7. Trends corresponding to melting of mantle are present at higher pressures (assuming that pressure sources with very small SC contributions (0 to <0.2%) does not play an important role in the dF/dSC curves, adequately reproduce variations in some OIB-like suites which is dicult to ascertain from existing models). (e.g., Dalles Lakes and Simcoe, Figs. 4, 7). Trends cor- Finally, we also examined the e€ects of di€erent com- responding to melting of mantle sources with very large positions of peridotite and SC on the trace element pre- SC contributions (>2%) also appear to produce melts dictions of isenthalpic ¯ux melting. In these examples, we with CAB compositions. However, trends corresponding use a strongly incompatible element depleted (MORB to melting of sources with intermediate SC contributions source) peridotite, and both SC compositions presented in (e.g., 0.5±2%) predict compositions that are intermedi- Table 4. ate between the observed CAB and OIB ®elds, where no samples are observed (Fig. 7). If trace element di€eren- Model results ces between CAB and OIB-like magmas were simply a matter of degree of SC metasomatism of each mantle Figure 7 shows ®elds of Cascade basalt compositions source, then a broad swath of compositions corre- with respect to LILE and HFSE contents (the ®elds sponding to a fan of positively correlated melting trends encompass the same data shown in Fig. 4). It also shows would be expected, not the clearly distinct and divergent the predictions of the ®rst two melting model scenarios: CAB-OIB ®elds (Figs. 4, 7). Simply varying the amount (1) progressive melting of individual metastomatized of SC in each mantle source (without correlation be-

Fig. 7 Fields of data shown in Melting of individual sources and isothermal flux melting Fig. 4, with model partial 80 melting trends of the SC-F 800 scenarios shown in Fig. 6A and a c B (progressive melting of indi- SC 5% 60 vidual sources, and isothermal 600 ¯ux melting). All partial melt- 1.5% SC 1225°C ing trends increase in melt fraction from right to left, and 1275°C 40 400 Ba are shown for up to 30% 1375°C Ba/Nb integrated melting. Arrows point in directions of increasing 0% SC degree of partial melting. 200 20 Dashed black lines are progres- sive melting of individual sources metasomatized by 0 0 either 0, 1.5, or 5% SC b d (Fig. 6A). Solid black lines are 3.0 melting trends for isothermal 8 SC-F ¯ux-melting (Fig. 6B), and are shown for peridotite 6 temperatures of 1225, 1275, and 2.0

1375 °C (colder mantle has Th Ba/Zr higher LILE and LILE/HFSE 4 at a given HFSE than hotter mantle). Neither melting of individual, variably metasoma- 2 1.0 tized sources nor isothermal ¯ux melting reproduce the 0 distinct and divergent CAB and 0 10 20 30 40 50 60 0.025 0.075 0.125 0.175 0.225 OIB ®elds in all plots Nb Nb/Zr

no SC-F correlation isothermal flux-melting 0%, 1.5%, 5% SC T: 1225, 1275, 1375°C 258 tween F and SC) cannot explain the CAB-OIB distinc- HFSE decrease. In the isenthalpic ¯ux melting models, tion. however, as F increases LILE concentrations reach a Trace element predictions of linear SC-F correlation minimum and then increase, while HFSE concentrations simulating isothermal ¯ux melting (e.g., the scenario continue to decrease or remain constant in progressively illustrated in Fig. 6B) predict partial melting trends higher degree melts. This transition occurs at about 10% (solid lines in Fig. 7) that reproduce the shape of the melting for an initial peridotite temperature of 1225 °C, data ®elds with respect to some trace element ratios and about 20% for 1375 °C (Fig. 8). This decoupling of (Fig. 7). However, these models do not predict the LILE and HFSE concentrations at high degrees of strong divergence of CAB and OIB-like magma com- melting occurs because SC-source metasomatism begins positions in trace element concentration space (Fig. 7), to have a large e€ect on LILE melt concentrations at and cannot produce the most extreme CAB magmas high degrees of melting (10±20%), whereas HFSE con- with the lowest HFSE and highest LILE. Isothermal ¯ux centrations are not signi®cantly a€ected by SC metaso- melting can only explain strongly CAB-like composi- matism. This decoupling means that high F melts of tions with both low HFSE and high LILE by melting relatively cold peridotite develop high LILE/HFSE, and (about 15±25%) extremely cold mantle (T < 1175 °Cin partial melting trends are V-shaped or hyperbolic and this model). Lower extents of melting (5±10%) of such diverge near the origin, similar to the distinct divergence cold mantle also predicts compositions intermediate of the Cascade CAB and OIB-like ®elds (Fig. 8). The between CAB and OIB-like magmas, where no samples shift from decreasing to increasing LILE concentrations are observed (Figs. 4, 7). Varying the peridotite and SC with increasing F is more pronounced for colder compositions within reasonable limits (e.g., Borg et al. peridotite because of the lower dF/dSC of colder mantle 1997 MORB to OIB compositions) also does not and thus larger SC contribution at a given F. Hotter produce a good ®t between the isothermal ¯ux melting mantle shows less of a LILE-HFSE divergence because trends and the CAB-OIB ®elds. We conclude that no SC addition produces a larger increase in F, thereby simple isothermal ¯ux melting model predicts the diluting LILE concentrations in resulting melts. strongly diverging and distinct CAB and OIB-like ®elds. In the isenthalpic ¯ux-melting model, low degree Trace element predictions of isenthalpic ¯ux melting melts (approximately <10%) are OIB-like regardless of (scenarios illustrated in Fig. 6C) reproduce many initial peridotite temperature, but if the initial tempera- features of the distinct and divergent CAB and OIB-like ture is lower than about 1250 °C (i.e., lower dF/dSC, or ®elds (Fig. 8). As observed in previous models, low a lower degree of melting per increment of SC metaso- degree (approximately <8%) partial melts have well matism), melts produced by degrees of melting greater correlated LILE and HFSE concentrations, and as both than about 10±15% show CAB trace-element signatures. SC and F increase, concentrations of both LILE and Higher initial mantle temperature (i.e., higher dF/dSC,

Fig. 8a±d Fields of data shown Isenthalpic flux melting in Fig. 4, with model partial 80 melting trends of isenthalpic 800 c ¯ux melting (scenario shown in a Fig. 6C) of peridotite with ini- tial temperatures of 1225, 1275, 60 600 and 1375 °C. All partial melt- 1225°C ing trends increase in melt fraction from right to left, and 40 400 1275°C Ba are shown for up to 30% Ba/Nb integrated melting. Arrows point in directions of increasing 1375°C degree of partial melting; circles 30% 20 200 5% are 5% integrated melting in- tervals. Isenthalpic ¯ux melting 15% trends reproduce most features 0 0 of the CAB-OIB distinction, b d and suggest that CAB signa- 3.0 tures can be explained as the 8 result of high degrees of melting (which dilutes incompatible elements), caused by high ex- 6 tents of SC metasomatism 2.0 Th (which enriches LILE but not 4 Ba/Zr HFSE in resulting melts)

2 1.0

0 0 10 20 30 40 50 60 0.025 0.075 0.125 0.175 0.225 Nb Nb/Zr 259 or a higher degree of melting per increment of SC that used by Borg et al. (1997) does not change the metasomatism) shifts the trend towards the origin, in patterns noticeably. which case even highly SC-metasomatized sources can- Figure 9B shows Ba and Nb concentrations of is- not produce signi®cantly LILE-enriched melts because enthalpic ¯ux-melts using the MORB source peridotite of the higher degree of melting accompanying SC composition of Borg et al (1997; Table 4). All melts in metasomatism (Fig. 8). This may indicate that magmas this example have extremely low Nb, re¯ecting the with CAB trace-element signatures can only be derived strong incompatible element depletion of the initial from melting of mantle regions with relatively low initial peridotite. Progressive SC metasomatism and melting temperatures. drive melt compositions to higher Ba concentrations at relatively low and constant Nb, due to the high Ba and low Nb content of the SC. This shows that E€ect of peridotite and SC compositions on isenthalpic-¯ux melting trends Effect of peridotite and SC composition Figure 9 illustrates the in¯uence of source peridotite and SC compositions on the trace element predictions of is- 800 enthalpic ¯ux-melting. In all cases, including the results in Fig. 8, trace-element trends record a shift from 1225 C peridotite dominated to SC-dominated compositions, 600 1275 C with progressive partial melting. The trends begin with 1375 C well-correlated concentrations of LILE and HFSE in low degree partial melts that decrease with progressive Ba 400 melting. With higher degrees of SC metasomatism and melting, and independent of the initial peridotite com- 5% position, the trends change direction, heading towards 30% the SC bulk composition. Compositional variation of 200 A the peridotite and SC sources can therefore have a large 15% Borg et al. e€ect on trace element trends of isenthalpic ¯ux melting. OIB-source Figure 9A shows Ba and Nb concentrations in ¯ux 800 melts using the same SC composition, but the slightly more Ba-enriched and Nb-depleted (Table 4) OIB- source peridotite composition of Borg et al. (1997), as in Fig. 8. The only signi®cant di€erence between Figs. 8 600 and 9A is the lower Nb concentrations in the initial low

degree partial melts in Fig. 9A. This suggests that the Ba higher Nb concentrations in our SC composition 400 (Figs. 5, 8) are necessary to produce the Cascade OIB- like lavas. Changing the SC composition in Fig. 9A to 1225 C

200 1275 C B c Borg et al. 1375 C Fig. 9A±C Model partial melting trends of isenthalpic ¯ux MORB-source melting, using di€erent peridotite and SC compositions. Arrows 0 point in directions of increasing degree of partial melting; circles are 5% integrated melting intervals. A Same SC composition as in Fig. 8, but OIB-source peridotite composition of Borg et al. (1997; Table 4). The lower Nb concentration of Borg et al. (1997) peridotite produces lavas with lower Nb than Cascade OIB-like 600 lavas. Other features of the results are similar to those in Fig. 8. Changing the SC composition to that used by Borg et al. does not

noticeably change the trends. B Isenthalpic ¯ux melting of Ba 1225 C peridotite, using same SC as in Fig. 8, but peridotite with a 400 strongly incompatible-element depleted MORB source composi- tion (Fig. 5; Borg et al. 1997). All melts have very low Nb, due to low Nb in both SC and peridotite. Higher degree melts of cold C mantle trend towards higher Ba, away from MORB-like compo- 200 1275 C sitions and towards, but at lower Nb than, CAB compositions. C Borg et al. Flux melting trends assuming SC and MORB-source compositions MORB-source of Borg et al. (1997). Results are similar to B, in that low-degree 1375 C and SC melts (and higher degree melts of hotter mantle) have compositions 0 similar to MORB-like melts, and higher degree melts of cold 0 10 20 30 40 50 60 mantle trend away from MORB, through CAB compositions, and towards the SC composition Nb 260 strongly incompatible element-depleted peridotite (i.e., peridotite and SC compositions; Table 4) that approxi- MORB source) cannot be the source of OIB-like mate the compositions of DL and TS magmas. The DL lavas. sample 96PRDR2 can be modeled as low degree melt (4± Using a MORB source peridotite composition, but 5%) of peridotite metasomatized by small amounts of with the higher Nb SC composition used by Borg et al. SC (0.2±0.3%). This corresponds to an isenthalpic ¯ux (1997; Table 4), shifts the higher degree melts towards melting trend with an initial peridotite temperature of higher Nb and Ba. This produces a positive correlation about 1325 °C (Figs. 6C, 8). In contrast, TS lavas can be of increasing Ba and Nb concentrations in larger degree modeled as larger degree melts (22%) of peridotite melts (Fig. 9C). One important result of these MORB metasomatized by large amounts of SC (4.0%), corre- source examples is that in both this and the previous sponding to an initial peridotite temperature of 1275 °C. example (Fig. 9B), CAB-like compositions can be Assuming an SC with 44% H2O (Stolper and Newman produced by relatively large extents of ¯ux melting of 1994), this would mean that the mantle sources of the relatively cold MORB source peridotite (>about 20% DL and TS lavas were ¯uxed by 0.1 and 1.8% H2O, for 1275 °C, and >about 10% for 1225 °C). The posi- corresponding to primary magma compositions with 2.2 tive correlation between Ba and Nb in the CAB ®eld in and 8.0% H2O, respectively. Fig. 9C is similar to variations observed in the Three Other studies have suggested similar relative and Sisters basalts (Fig. 4), suggesting that these lavas could absolute extents of melting for similar types of arc be produced by ¯uxing of a less Nb-depleted SC com- basalts. Stolper and Newman (1994) concluded that position similar to that used in Borg et al. (1997). LILE/HFSE, H2O content, and degree of melting were In summary, variations in SC composition, and correlated in Mariana back arc (F  5±20%) and arc especially peridotite composition, can have strong (F  15±35%) lavas. Baker et al. (1994) also found in¯uences on the trace element patterns of ¯ux-melting. similar results from southern Cascades (Mt. Shasta) The peridotite and SC compositions used in the models lavas, and in combination with experimental evidence shown in Fig. 8 can reproduce nearly the entire ®eld of suggested that high alumina basalts have low H2O CAB and OIB-like Cascade magmas, as well as MORB- (essentially anhydrous), and were produced by low like lavas with low incompatible element concentrations. extents of melting (6±10%), whereas high MgO basaltic However, Fig. 9 also shows that CAB and MORB-like andesites and andesites have high H2O (3±6%) and were magmas (but not OIB-like magmas) can also be produced by higher degrees of melting (20±30%). In produced by ¯ux melting of a strongly incompatible agreement with the interpretations here, the studies of element-depleted MORB source peridotite. Stolper and Newman (1994) and Baker et al. (1994) suggest that LILE/HFSE, H2O, and degree of melting are generally correlated in primitive arc magmas, and Isenthalpic ¯ux-melting and Cascade basalts re¯ect the ¯ux melting induced correlation between SC addition and degree of melting. This contradicts the Figure 10 shows trace element compositions of model interpretations of Conrey et al. (1997) that degree of melts produced by isenthalpic ¯ux melting (using our melting decreases for magmas with progressively higher LILE/HFSE and H2O (in the order OIB-like basalt, Fig. 10 Trace-element compositions of model melts produced by CAB, basaltic andesite, and absarokite). Conrey et al.'s isenthalpic ¯ux melting, compared with compositions of DL and (1997) inferred inverse relationship between LILE/ TS basalts. Assuming our peridotite and SC compositions shown in HFSE and degree of melting also requires a special Fig. 5 and a ¯ux-melting relation of Eq. (3), DL magmas can be explanation for the strong HFSE depletions in basaltic adequately modeled by relatively low F and SC, with an initial peridotite source temperature of about 1325 °C, whereas TS andesites and absarokites, in the form of residual HFSE- magmas can be modeled by relatively high F and SC, with an bearing minerals, unusual mantle source, or melt initial peridotite source temperature of about 1275 °C peridotite reaction. In contrast, ¯ux melting, with the

100 ve mantle ve 10 Dalles Lakes

Dalles Lakes model (SC = 0.22%, F = 4.5%), T0 = 1325°C

Sample/primiti Three Sisters

Three Sisters model (SC = 4.0%, F = 22%), T0 = 1275°C 1.0 Cs Rb Ba Th U Nb Ta K La Ce Sr Nd Sm Zr Hf Eu Ti Gd Yb 261 main feature of correlated extents of HFSE depletion Isotope-trace element correlation (by high degrees of melting), and LILE and H2O en- richment (by SC metasomatism), explains the main Although most isotopic and trace element compositions compositional features of these diverse types of primitive of primitive arc basalts from southern Washington and arc magmas. It should also be noted that especially high northern Oregon are not well correlated, Ba/Nb and LILE/HFSE magmas, such as absarokites, may not re- 87Sr/86Sr do show some correlation (Leeman et al. 1990; quire extremely high degrees of melting in any individual Conrey, unpublished data; Fig. 11). By assuming a melting event. Instead, they may represent moderate peridotite 87Sr/86Sr of 0.7027, an SC 87Sr/86Sr 0.7036± degrees of ¯ux melting of peridotite depleted by previous 0.7038, and modeling melts as isenthalpic ¯ux melts ¯ux melting events, as proposed for the Mariana arc from peridotite at a range of temperatures between magmas by Stolper and Newman (1994). approximately 1225 and 1350 °C, we can reproduce the 87 86 The di€erences in predicted H2O contents of the overall Ba/Nb and Sr/ Sr correlation (Fig. 11). This is primary DL and TS lavas from the isenthalpic ¯ux not a unique solution but the correspondence of the melting model are consistent with evidence from other isotopic-trace element correlation to the prediction of studies, but the predicted H2O content of the TS the ¯ux melting trends supports the proposal of an SC-F magma is high relative to estimates for most arc melts. correlation, and a dominant role of ¯ux melting in Based on petrologic constraints and comparison to Cascade arc magma genesis. experimental evidence, OIB-like basalts from the Cas- cades have been estimated to have low to moderate amounts of H2O (e.g., up to 1%; Conrey et al. 1997). Origin of the CAB trace-element signature In contrast, on the basis of melt inclusion composi- tions, petrographic and petrologic estimates, and ex- Our interpretations of a ¯ux melting cause for trace perimental evidence (Anderson 1974a, b, 1976; Sisson element di€erences between CAB and OIB-like basalts and Grove 1993; Baker et al. 1994), several studies have support previous interpretations of a correlation between estimated magmas parental to basaltic andesites and ¯uid ¯uxing and degree of melting in some continental arc andesites with CAB signatures from Mt. Shasta to magmas (e.g., Barragan et al. 1998). In addition, our contain 3±6.5% H2O. Other measurements of H2O modeling supports previous suggestions (Gill 1981; contents in ma®c melt inclusions in arc basalts and basaltic andesites range widely from nearly anhydrous for high alumina MORB-like basalts, to >6 wt% in 0.7037 some lavas, with typical values of about 2% (e.g., a Harris and Anderson 1984; Sisson and Layne 1993; Sobolev and Chaussidon 1996). 25 0.7035 20 The prediction of 8% H2O for the primary Three 15 Sisters magma is high compared with these measure- b ments. Assuming that this prediction is too high for a 0.7033 10 primary arc magma, there are a number of possible reasons for this discrepancy. These include: (1) overes- Sr/ Sr timation of the SC H O content, (2) decoupling of H O 87 86 2 2 0.7031 5 from trace elements during SC in®ltration, melting, or melt migration, (3) incomplete H2O extraction from the SC metasomatized source, possibly because of the 0.7029 presence of residual hydrous phases (which may mean amphibole or other hydrous phases are required in trace element models), or (4) a multiple-stage melting process 0.7027 for the TS lavas such that the 22% estimate is an inte- 5254565 grated value re¯ecting more than one SC ¯uxing and Ba/Nb melt depletion event (e.g., Stolper and Newman 1994). Fig. 11 Ba/Nb and 87Sr/86Sr of basalts from Mt. Adams, Mt. St. In any case, comparing and reconciling H2O contents Helens, Simcoe, and Indian Heaven (data from Leeman et al. 1990; and trace element signatures of primary arc magmas is Bacon et al. 1997) and the northern Oregon Cascades (Conrey, an important issue for the evaluation of ¯ux melting unpublished data) with compositions of model melts for isenthalpic models. The relatively (and probably unrealistically) ¯ux-melting. Model trends assume our peridotite and SC compo- sitions (Table 4), an initial peridotite temperature of 1275 °C, and high H2O contents implied by this model for strongly peridotite 87Sr/86Sr of 0.7027, and SC 87Sr/86Sr of 0.7038 (trend a) CAB lavas such as the Three Sisters basalts may suggest and 0.7036 (trend b). Using mantle temperatures from about 1225 additional complications in the subarc melting process. to 1375 °C makes little di€erence in the position of the model One possible example is chromatographic e€ects on the trend, although the positions of integrated melt fractions along the trends do change. Three northern Oregon samples with Ba/Nb SC or melt (e.g., Stolper and Newman 1994) that may greater than 75 have been cut o€ to show detail in the lower Ba/Nb preclude using a single SC composition to model all part of the curve, but these three samples have high 87Sr/86Sr melts. (0.7035±0.7037) that fall on projections of the melting trends 262 Woodhead 1989; Stolper and Newman 1994) that the Leeman et al. 1990). The trace element models presented distinctive negative HFSE anomalies and high LILE/ here, however, indicate that OIB-like arc lavas do not HFSE of CAB magmas can be explained simply by large necessarily preclude the addition of slab-derived SC to extents of SC metasomatism and large degrees of melting the mantle wedge, or magmatism by ¯ux melting. Lavas of mixed peridotite-SC source, and do not require special with a range of trace element signatures, from CAB to residual Nb- and Ta-rich phases. Alone, high extent of OIB-like (with a distinctive divergence in trace element melting produces melts with low incompatible element plots as in Fig. 4), can be simply produced by a single concentrations, but the correlation (and probably causal process of SC-induced ¯ux melting of a single mantle relationship between) the extent of SC metasomatism and source composition over a range of mantle tempera- degree of melting means that as HFSE concentrations tures. It is important to note however, that OIB-like arc decrease due to progressive melting, LILE concentrations basalts cannot be produced from strongly incompatible are simultaneously enriched in the source and potentially, element-depleted mantle sources (i.e., MORB source in the melt. The important question for the origin of the mantle) by ¯ux melting, but CAB lavas can. CAB signature is whether dilution by melting or source Other workers (e.g., Ma rquez et al. 1999) have sug- enrichment by SC metasomatism more strongly in¯u- gested that in some cases OIB-like magmas in an arc ences melt compositions. Two conditions are necessary to setting indicate the presence of a mantle plume beneath produce the high LILE/HFSE characteristic of CAB the arc. Our results, however, clearly show that the trace magmas: low dF/dSC and high F. If dF/dSC is low, LILE element signatures that typify OIB-like arc lavas can be enrichment by SC metasomatism overwhelms LILE de- produced simply by relatively low degrees of a peridotite pletion by melting. If, in addition, the degree of melting is source that is not strongly depleted of incompatible high so that HFSE concentrations are depleted, the re- elements. This could occur by partial melting either in sulting magma will have high LILE/HFSE. These e€ects response to SC ¯uxing or not, but the common associ- are most pronounced for high degrees of melting of cold ation of CAB and OIB-like lavas in arcs suggests that peridotite, and isenthalpic ¯ux melting of such mantle can ¯ux melting plays at least some role. produce CAB magmas regardless of its initial composi- In summary, our results have shown that OIB-like tion (Figs. 8, 9). magmas in arcs do not necessarily preclude the addition Several di€erent scenarios can lead to OIB- or of slab-derived SC in the mantle wedge, nor do they MORB-like melt compositions instead of CAB. Firstly, require special mantle plume activity under an arc. if the peridotite is initially cold (low dF/dSC), but the Because CAB trace element signatures require both low degree of melting is low, then LILE and HFSE con- dF/dSC and high F, OIB-like (or MORB-like) magmas centrations in the melt are not strongly decoupled and can be expected and explained in the context of arc magmas do not develop a CAB signature. In this case, if melting, in cases where dF/dSC is high, or F is low, or the initial peridotite is not strongly depleted of incom- both. In the absence of robust isotopic constraints (such patible elements, the melts are OIB-like (Fig. 8), whereas as He isotopes) indicative of plume-like mantle sources, if the initial peridotite is strongly depleted of incom- it is simpler to explain OIB-like arc magmas by the same patible elements, the melts are MORB-like (Fig. 9B, C). ¯ux melting process and the same peridotite and SC Secondly, if the degree of melting is high, but the mantle sources that produce CAB magmas. A similar interpr- source is initially hot (high dF/dSC), melting keeps pace etation could be warranted in cases of older volcanic with SC metasomatism so that addition of LILE has rocks for which the tectonic environments are not clear. little e€ect compared with dilution by melting, and LILE Lack of CAB-like trace element signatures (and the and HFSE concentrations are both low. In this case, if presence of OIB- or MORB-like compositions) in an- the peridotite is not depleted and the degree of melting is cient lava sequences may not necessarily preclude an arc not too high, the melts have an OIB-like composition environment for their formation, especially given that (Fig. 8), whereas if the peridotite is initially depleted or mantle wedge temperatures (and therefore dF/dSC) were the degree of melting is high, the melts are MORB-like probably higher in Precambrian arcs. in composition (Fig. 9B, C). Finally, if the degree of melting is low, but the mantle is initially hot, melts can be either OIB- or MORB-like, depending on the initial Conclusions peridotite composition (Figs. 8, 9). The young ages of the DL and Three Sisters basalts (2.2 Ma and 370±400 ka, respectively) show that not all The signi®cance of OIB-like magmas in arcs the late Cenozoic to recent volcanism in the central Washington Cascades is associated with large andesitic The observation that many primitive Cascade magmas stratocones and their satellitic vents. The compositions possess OIB-like, rather than CAB trace element signa- of basalts from these two vents are distinct in several tures, has led to the suggestion that material from the respects. Most importantly, they have strongly con- subducted slab plays little role in Cascade petrogenesis, trasting LILE/HFSE that characterize the CAB-OIB and that other processes such as shear heating and distinction for many Cascade lavas, as well as lavas from mantle convection play a more dominant role (e.g., other arcs. Simple predictions of ¯ux melting of mantle 263 peridotite can adequately reproduce the strongly diver- varying melt fraction and water content. Contrib Mineral Pet- gent trace element compositions of both CAB and OIB- rol 118: 111±129 Barragan R, Geist D, Hall M, Larson P, Kurz M (1998) Sub- like magmas. Correlation between the amount of sub- duction controls on the compositions of lavas from the Ecua- duction-related ¯uid (SC) added to a peridotite source, dorian Andes. Earth Planet Sci Lett 154: 153±166 and the degree of melting of the source, produces trace Borg LE, Clynne MA, Bullen TD (1997) The variable role of slab- element trends that appear to form diverging arrays, not derived ¯uids in the generation of a suite of primitive calc- predicted by melting of a continuum of individual alkaline lavas from the southernmost Cascades, California. Can Mineral 35: 425±452 sources. The unusual trace element trends of ¯ux melting Conrey RM, Sherrod DR, Hooper PR, Swanson DA (1997) are due to the depletion of incompatible elements in the Diverse primitive magmas in the Cascade arc, northern Oregon melt with increasing degree of melting, and simultaneous and southern Washington. Can Mineral 35: 367±396 enrichment of LILE by SC addition to the source. If Dunn T, Stolper EM, Baker MB (1993) Partial melting of an amphibole-bearing lherzolite at 1.5 GPa (Abstr). Geol Soc Am both the amount of SC addition and the degree of Abstr with Prog 25: 213 melting are suciently high, magmas will have low Fischer JF (1970) The geology of the White River-Carbon Ridge HFSE and high LILE, characteristic CAB signatures. area, Cedar Lake quadrangle, Cascade Mountains, Washing- This should be the case for large degrees of isenthalpic ton. PhD Thesis, University of California, Santa Barbara Gaetani GA, Grove TL (1998) The in¯uence of water on melting of ¯ux melting of relatively cold mantle. Other situations, mantle peridotite. Contrib Mineral Petrol 131: 323±346 such as low degrees of melting or low amounts of SC Gill JB (1981) Orogenic andesites and plate tectonics. Springer, addition relative to degree of melting, will produce OIB- Berlin Heidelberg New York, 358 pp and MORB-like magmas. Hammond PE (1980) Reconnaissance geologic map (1:125,000) and It is likely that the extents of melting and SC meta- cross sections of southern Washington Cascade Range, latitude 45°30¢N±47°15¢N, longitude 120°45¢±122°22.5¢W: Portland somatism are often correlated in arc magmatism. Given State University, Department of Earth Sciences, p 31, 2 pl that SC has large concentrations of many incompatible Hammond PE (1998) Tertiary andesitic lava-¯ow complexes elements, and as shown here by the complicated trace (stratovolcanoes) in the southern Cascade Range of Washing- element trajectories of ¯ux melting, it will be dicult to ton ± observations on tectonic processes within the Cascade arc. Wash Geol 26: 20±30 accurately interpret magmatic compositional variations Harris DM, Anderson AT Jr (1984) Volatiles H2O, CO2, and Cl in or determine mantle source compositions without ex- a subduction related basalt. Contrib Mineral Petrol 87: 120±128 plicitly modeling this correlation. Other types of mag- Hartman DA (1973) Geology and low-grade metamorphism of the mas may also have an origin involving ¯ux melting, Greenwater River area, central Cascade Range, Washington. PhD Thesis, University of Washington, Seattle especially those not obviously related to mantle upwel- Hawkesworth CJ, Gallagher K, Hergt JM, McDermott F (1993) ling beneath ridges or hotspots. In such cases, mantle Mantle and slab contributions in arc magmas. Annu Rev Earth metasomatism that induces melting may be analogous to Planet Sci 21: 175±204 SC addition in arcs, and the possible link between Hawkesworth CJ, Gallagher K, Hergt JM, McDermott F (1994) mantle source composition and degree of melting should Destructive plate margin magmatism: geochemistry and melt generation. Lithos 33: 169±188 be considered. Hess PC (1992) Phase equilibria constraints on the origin of ocean ¯oor basalts. In: Phipps Morgan J, Blackman DK, Sinton JM Acknowledgements We thank Grant Newport and Jeannie Hunt of (eds) Mantle ¯ow and melt generation at mid-ocean ridges. Am Weyerhauser for access to and navigational help in the Three Sis- Geophy Union Geophys Monogr 71: 67±102 ters region. We appreciate constructive reviews from Jim Luhr and Hirschmann MM, Asimow PD, Ghiorso MS, Stolper EM (1999) Rick Conrey, and comments on an early version of this paper by Calculation of peridotite partial melting from thermodynamic Gwyneth Jones and Bruce Nelson. We also appreciate helpful models of minerals and melts. III. Controls on isobaric melt discussions with John Eiler, Marc Hirschmann, and Tom Sisson. production and the e€ect of water on melt production. J Petrol Thanks to Ken Farley for support for Reiners' moonlighting on 40: 831±851 this project at Caltech. Hirose K, Kawamoto T (1995) Hydrous partial melting of lherzolite at 1GPa ± the e€ect of H2O on the genesis of basaltic magmas. Earth Planet Sci Lett 133: 463±473 Ionov DA, Hofmann AW (1995) Nb-Ta-rich mantle amphiboles References and micas: Implications for subduction-related metasomatic trace element fractionations. Earth Planet Sci Lett 131: 341± Anderson AT (1974a) Evidence for a picritic, volatile-rich magma 356 beneath Mt. Shasta, California. J Petrol 15: 243±267 Langmuir CH, Klein EM, Plank T (1992) Petrological systematics Anderson AT (1974b) Chlorine, sulfur, and water in magmas and of mid-ocean ridge basalts: constraints on melt generation be- oceans. Geol Soc Am Bull 85: 1492±1585 neath ocean ridges. In: Phipps Morgan J, Blackman DK, Sin- Anderson AT (1976) Magma mixing: Petrological process and ton JM (eds) Mantle ¯ow and melt generation at mid-ocean volcanological tool. J Volcanol Geothermal Res 1: 3±33 ridges. Am Geophys Union Geophys Monogr 71: 183±280 Ayers JC, Dittmer SK, Layne GD (1997) Partitioning of elements Lanphere MA, Sisson TW (1995) K-Ar ages of Mount Rainier between peridotite and H2O at 2.0±3.0 GPa and 900±1100 °C, volcanics. EOS Trans Am Geophy Union 76: 651 and application to models of subduction zone processes. Earth Leeman WP, Smith DR, Hildreth W, Palacz Z, Roges N (1990) Planet Sci Lett 150: 381±398 Compositional diversity of late Cenozoic basalts in a transect Bacon CR, Bruggman PE, Christiansen RL, Clynne MA, Don- across the southern Washington Cascades: implications nelly-Nolan JM, Hildreth W (1997) Primitive magmas at ®ve for subduction zone magmatism. J Geophys Res 95: 19561± Cascade volcanic ®elds: melts from hot, heterogeneous sub-arc 19582 mantle. Can Mineral 35: 397±423 Luedke RG, Smith RL (1982) Map (1:1,000,000) showing the Baker MB, Grove TL, Price R (1994) Primitive basalts and ande- distribution, composition, and age of late Cenozoic volcanic sites from the Mt. Shasta region, N. California: products of centers in Oregon and Washington. USGS Map I-1091-D 264

Luhr JF (1997) Extensional tectonics and the diverse primitive Sisson TW, Lanphere MA (1997) The growth of Mount Rainier volcanic rocks in the Western Mexican volcanic belt. Can volcano, Cascade arc, USA. IAVCEI General Assembly, Mineral 35: 473±500 Puerto Vallarta, Mexico, p 5 Luhr JF, Allan JF, Carmichael ISE, Nelson SA, Hasenaka T (1989) Sisson TW, Layne GD (1993) H2O in basalt and basaltic andesite Primitive calc-alkaline and alkaline rock types from the western glass inclusions from four subduction-related volcanoes. Earth Mexican volcanic belt. J Geophys Res 94: 4515±4530 Planet Sci Lett 117: 619±635 Ma rquez A, Oyarzun R, Doblas M, Verma SP (1999) Alkalic Sobolev AV, Chaussidon M (1996) H2O concentrations in primary (ocean-island basalt type) and calc-alkalic volcanism in the melts from supra-subduction zones and mid-ocean ridges: im- Mexican volcanic belt: a case for plume-related magmatism plications for H2O storage and recycling in the mantle. Earth and propagating rifting at an active margin? Geology 27: Planet Sci Lett 137: 45±55 51±54 Stolper E, Newman S (1994) The role of water in the petrogenesis McKenna JM (1994) Geochemistry and petrology of Mount of Mariana trough magmas. Earth Planet Sci Lett 121: 293± Rainier magmas: petrogenesis at an arc-related stratovolcano 325 with multiple vents. MS Thesis, University of Washington, Sun M, Kerrich R (1995) Rare earth element and high ®eld strength Seattle element characteristics of whole rocks and mineral separates of Menner AV, Dunn T (1995) Amphibole and phlogopite stability in ultrama®c nodules in Cenozoic volcanic vents of southeastern an initially amphibole-bearing spinel peridotite under water- British Columbia, Canada. Geochim Cosmochim Acta 59: saturated conditions; 1 to 2.5 GPa (Abstr). EOS Trans Am 4863±4879 Geophys Union 76 Suppl 46: 697 Sun SS, McDonough WF (1989) Chemical and isotopic Miyashiro A (1978) Nature of alkalic volcanic rock series. Contrib systematics of oceanic basalts: implications for mantle com- Mineral Petrol 66: 91±104 position and processes. In: Saunders AD, Norry MJ (eds) Niida K, Green DH (1999) Stability and chemical composition of Magmatism in the ocean basins. Blackwell Scienti®c, London, pargasitic amphibole in MORB pyrolite under upper mantle pp 313±345 conditions. Contrib Mineral Petrol 135: 18±40 Tatsumi Y, Hamilton DL, Nesbitt RW (1986) Chemical charac- Sen C, Dunn T (1995) Experimental modal metasomatism of a teristics of ¯uid phase released from a subducted lithosphere and spinel lherzolite and the production of amphibole-bearing origin of arc magmas: evidence from high-pressure experiments peridotite. Contrib Mineral Petrol 119: 422±432 and natural rocks. J Volcanol Geotherm Res 29: 293±309 Sisson TW, Grove TL (1993) Temperature and H2O contents of low- Woodhead JD (1989) Geochemistry of the Mariana arc (western MgO high-alumina basalts. Contrib Mineral Petrol 113: 167±184 Paci®c): source composition and processes. Chem Geol 76: 1±24